Page 53 of 117 FirstFirst ... 3132333435051525354555663738393103 ... LastLast
Results 833 to 848 of 1860

Thread: AMD - Zen chitchat

  1. #833
    Moosekateer CAT-THE-FIFTH's Avatar
    Join Date
    Aug 2006
    Location
    Not here
    Posts
    29,497
    Thanks
    3,318
    Thanked
    4,627 times in 3,571 posts
    • CAT-THE-FIFTH's system
      • Motherboard:
      • Less E-PEEN
      • CPU:
      • Massive E-PEEN
      • Memory:
      • RGB E-PEEN
      • Storage:
      • Not in any order
      • Graphics card(s):
      • EVEN BIGGER E-PEEN
      • PSU:
      • OVERSIZED
      • Case:
      • UNDERSIZED
      • Operating System:
      • DOS 6.22
      • Monitor(s):
      • NOT USUALLY ON....WHEN I POST
      • Internet:
      • FUNCTIONAL

    Re: AMD - Zen chitchat

    Quote Originally Posted by DanceswithUnix View Post
    That would be odd.

    Die size of Summit Ridge/Zepplin is 192 mm^2, die size of Raven Ridge APU is 210 mm^2.

    So it would be cheaper to heavily fuse off the old Ryzen cpu. ThreadTickler?
    Why would it be when it's been the case back to Llano?? After all there must be APUs with failed IGPs??


    Those despicable Elk,stealing the pond weed!

  2. #834
    root Member DanceswithUnix's Avatar
    Join Date
    Jan 2006
    Location
    In the middle of a core dump
    Posts
    10,808
    Thanks
    577
    Thanked
    1,160 times in 986 posts
    • DanceswithUnix's system
      • Motherboard:
      • Asus X470-PRO
      • CPU:
      • 3700X
      • Memory:
      • 16GB 3200MHz
      • Storage:
      • 1TB Linux, 1TB Games (Win 10)
      • Graphics card(s):
      • Asus Strix RX Vega 56
      • PSU:
      • 650W Corsair TX
      • Case:
      • Antec 300
      • Operating System:
      • Fedora 30 + Win 10 Pro 64 (yuk)
      • Monitor(s):
      • Benq XL2730Z 1440p + Samsung 2343BW 2048x1152
      • Internet:
      • Zen 80Mb/20Mb VDSL

    Re: AMD - Zen chitchat

    Quote Originally Posted by CAT-THE-FIFTH View Post
    Why would it be when it's been the case back to Llano?? After all there must be APUs with failed IGPs??
    There will, but I wouldn't expect enough to base a product off that. Most likely is a defect in one of the CUs which make up most of the area and can be disabled leaving a lower end working part, the critical support circuitry like video transcode doesn't look that big.

    I suspect most of the FM2 etc Athlons are just fused off to create a product rather than salvaged parts.

    Now if Raven ridge doesn't sell so well and AMD have a warehouse full of inventory, then fusing some off as Athlons could be a valid business decision. Otherwise, Summit Ridge and raven Ridge are both equally valid ways to make an Athlon X2/X4, but one is cheaper to make.

  3. #835
    Moosekateer CAT-THE-FIFTH's Avatar
    Join Date
    Aug 2006
    Location
    Not here
    Posts
    29,497
    Thanks
    3,318
    Thanked
    4,627 times in 3,571 posts
    • CAT-THE-FIFTH's system
      • Motherboard:
      • Less E-PEEN
      • CPU:
      • Massive E-PEEN
      • Memory:
      • RGB E-PEEN
      • Storage:
      • Not in any order
      • Graphics card(s):
      • EVEN BIGGER E-PEEN
      • PSU:
      • OVERSIZED
      • Case:
      • UNDERSIZED
      • Operating System:
      • DOS 6.22
      • Monitor(s):
      • NOT USUALLY ON....WHEN I POST
      • Internet:
      • FUNCTIONAL

    Re: AMD - Zen chitchat

    Quote Originally Posted by DanceswithUnix View Post
    There will, but I wouldn't expect enough to base a product off that. Most likely is a defect in one of the CUs which make up most of the area and can be disabled leaving a lower end working part, the critical support circuitry like video transcode doesn't look that big.

    I suspect most of the FM2 etc Athlons are just fused off to create a product rather than salvaged parts.

    Now if Raven ridge doesn't sell so well and AMD have a warehouse full of inventory, then fusing some off as Athlons could be a valid business decision. Otherwise, Summit Ridge and raven Ridge are both equally valid ways to make an Athlon X2/X4, but one is cheaper to make.
    Raven Ridge will be the higher volume part though. Its why AMD had two seperate dies last generation for their APUs.


    Those despicable Elk,stealing the pond weed!

  4. #836
    Not a good person scaryjim's Avatar
    Join Date
    Jan 2009
    Location
    Manchester
    Posts
    15,149
    Thanks
    1,224
    Thanked
    2,273 times in 1,865 posts
    • scaryjim's system
      • Motherboard:
      • Dell Inspiron
      • CPU:
      • Core i5 8250U
      • Memory:
      • 1x 8GB DDR4 2400
      • Storage:
      • 128GB M.2 SSD + 1TB HDD
      • Graphics card(s):
      • Radeon R5 230
      • PSU:
      • Battery/Dell brick
      • Case:
      • Dell Inspiron 5570
      • Operating System:
      • Windows 10
      • Monitor(s):
      • 15" 1080p laptop panel

    Re: AMD - Zen chitchat

    Quote Originally Posted by CAT-THE-FIFTH View Post
    Raven Ridge will be the higher volume part though. Its why AMD had two seperate dies last generation for their APUs.
    I'm not sure it will this time around. Don't forget that Ryzen CPUs, Threadripper and EPYC all use the same die. They must be churning those out by the absolute bucketload (in fact, they're fabbing so many they can afford to use the failed dies as packaging material in Threadripper).

    With the 2C/3CU Ryzen 3 mobile looking like a salvage part from Raven Ridge, they've got a lot of scope for fusing/binning for APUs. With Ryzen branding across the board, I don't see how they'd justify resurrecting the Athlon brand. I suspect it's rather more likely that we'll see a Ryzen 3 2100G in the sub $99 space, with 2 cores and 3CUs, salvaged from Raven Ridge.

    We had cheap Athlon X4s previously because they were the only way for AMD to compete with Intel - their dual-core APUs cost the same as a Pentium dual core but got stomped performance-wise; they needed the competitive advantage of more cores at the same price. That's no longer necessary - a dual core with SMT will fill the market nicely, and with 11 CUs on the die they're far more likely to be able to harvest parts with working IGPs. The 2200G is already cheaper than the 1300X - chop the IGP off and sell it even cheaper than that and you're just cannibalising your own sales - when the 1300X is cheaper to make!

    With a 3CU target IGP and 8 CUs on the die they needed 37.5% of them functional. Now they only need 27% functional. The tolerance for silicon defects is quite a lot higher. As I said earlier, the only circumstance I could see quad core Raven Ridge CPUs under is if they can maintain CPU clocks at lower TDPs, so they can be sold as energy efficient parts. But they wouldn't be cheaper than the APUs/Summit Ridge CPUs, I doubt...

  5. #837
    Moosekateer CAT-THE-FIFTH's Avatar
    Join Date
    Aug 2006
    Location
    Not here
    Posts
    29,497
    Thanks
    3,318
    Thanked
    4,627 times in 3,571 posts
    • CAT-THE-FIFTH's system
      • Motherboard:
      • Less E-PEEN
      • CPU:
      • Massive E-PEEN
      • Memory:
      • RGB E-PEEN
      • Storage:
      • Not in any order
      • Graphics card(s):
      • EVEN BIGGER E-PEEN
      • PSU:
      • OVERSIZED
      • Case:
      • UNDERSIZED
      • Operating System:
      • DOS 6.22
      • Monitor(s):
      • NOT USUALLY ON....WHEN I POST
      • Internet:
      • FUNCTIONAL

    Re: AMD - Zen chitchat

    Quote Originally Posted by scaryjim View Post
    I'm not sure it will this time around. Don't forget that Ryzen CPUs, Threadripper and EPYC all use the same die. They must be churning those out by the absolute bucketload (in fact, they're fabbing so many they can afford to use the failed dies as packaging material in Threadripper).

    With the 2C/3CU Ryzen 3 mobile looking like a salvage part from Raven Ridge, they've got a lot of scope for fusing/binning for APUs. With Ryzen branding across the board, I don't see how they'd justify resurrecting the Athlon brand. I suspect it's rather more likely that we'll see a Ryzen 3 2100G in the sub $99 space, with 2 cores and 3CUs, salvaged from Raven Ridge.

    We had cheap Athlon X4s previously because they were the only way for AMD to compete with Intel - their dual-core APUs cost the same as a Pentium dual core but got stomped performance-wise; they needed the competitive advantage of more cores at the same price. That's no longer necessary - a dual core with SMT will fill the market nicely, and with 11 CUs on the die they're far more likely to be able to harvest parts with working IGPs. The 2200G is already cheaper than the 1300X - chop the IGP off and sell it even cheaper than that and you're just cannibalising your own sales - when the 1300X is cheaper to make!

    With a 3CU target IGP and 8 CUs on the die they needed 37.5% of them functional. Now they only need 27% functional. The tolerance for silicon defects is quite a lot higher. As I said earlier, the only circumstance I could see quad core Raven Ridge CPUs under is if they can maintain CPU clocks at lower TDPs, so they can be sold as energy efficient parts. But they wouldn't be cheaper than the APUs/Summit Ridge CPUs, I doubt...
    They are the higher volume part though as AMD server share is tiny and even if it tripled it would still be tiny. The DIY build market is also tiny compared to the combined laptop and prebuilt desktop market. Why do you think AMD is having such issues with its consumer dGPU market?? Its been utterly knocked out of the laptop market by Nvidia.

    Also,remember this - GF has traditionally struggled to build GPUs to the same level as TSMC. The reason why we had the Athlon X4 CPUs derived from APUs,since in many cases the GPU sections were faulty. Look at the missed clockspeeds on the GPU section for things like Kaveri,etc or the fact Polaris had a huge amount of revisions.

    If anything why did we have the Excavator based Athlon II X4 845 on FM2 in the first place?? Its because AMD probably had loads of faulty Carrizo based APUs which had issues with their IGPs. It was a part which was not needed especially since Steamroller based Athlon II X4 CPUs which were actually slightly smaller existed.

    If anything I expect the Ryzen CPUs to simply have better yields,so it makes more sense for AMD to not artificially cripple them when they will have faulty Raven Ridge APUs they can sell instead.

    Edit!!

    Also with a 9% difference in surface area,I think its not enough to really affect margins that much.

    Second Edit!!

    Plus I expect the desktop Raven Ridge APUs to be failed laptop SKUs which are more leakier which is probably fine for desktop.

    It would not surprise me one bit,once Ryzen+ is released we might see a more simplified line-up,ie, Raven Ridge below £150 and Ryzen+ starting with a higher clockspeed 4C/8T version.
    Last edited by CAT-THE-FIFTH; 10-01-2018 at 02:37 PM.


    Those despicable Elk,stealing the pond weed!

  6. #838
    root Member DanceswithUnix's Avatar
    Join Date
    Jan 2006
    Location
    In the middle of a core dump
    Posts
    10,808
    Thanks
    577
    Thanked
    1,160 times in 986 posts
    • DanceswithUnix's system
      • Motherboard:
      • Asus X470-PRO
      • CPU:
      • 3700X
      • Memory:
      • 16GB 3200MHz
      • Storage:
      • 1TB Linux, 1TB Games (Win 10)
      • Graphics card(s):
      • Asus Strix RX Vega 56
      • PSU:
      • 650W Corsair TX
      • Case:
      • Antec 300
      • Operating System:
      • Fedora 30 + Win 10 Pro 64 (yuk)
      • Monitor(s):
      • Benq XL2730Z 1440p + Samsung 2343BW 2048x1152
      • Internet:
      • Zen 80Mb/20Mb VDSL

    Re: AMD - Zen chitchat

    Quote Originally Posted by CAT-THE-FIFTH View Post
    They are the higher volume part though...
    I was talking raw die cost, which AIUI is down to a simple "you pay for a wafer start, how many good die you get from that is your problem". Popularity/volume washes out of the equation, if mobile parts are really popular driving down the cost of raw wafers then Threadripper uses the same raw wafers so benefits just as much. In terms of yield, 9% extra area sounds pretty significant to me.

    And as I said, I don't believe for a second that Athlon X4 FM2 chips have faulty video circuitry. They are fused off and the GPU power budget aimed back at the CPU to create a new product.

  7. #839
    Moosekateer CAT-THE-FIFTH's Avatar
    Join Date
    Aug 2006
    Location
    Not here
    Posts
    29,497
    Thanks
    3,318
    Thanked
    4,627 times in 3,571 posts
    • CAT-THE-FIFTH's system
      • Motherboard:
      • Less E-PEEN
      • CPU:
      • Massive E-PEEN
      • Memory:
      • RGB E-PEEN
      • Storage:
      • Not in any order
      • Graphics card(s):
      • EVEN BIGGER E-PEEN
      • PSU:
      • OVERSIZED
      • Case:
      • UNDERSIZED
      • Operating System:
      • DOS 6.22
      • Monitor(s):
      • NOT USUALLY ON....WHEN I POST
      • Internet:
      • FUNCTIONAL

    Re: AMD - Zen chitchat

    Quote Originally Posted by DanceswithUnix View Post
    I was talking raw die cost, which AIUI is down to a simple "you pay for a wafer start, how many good die you get from that is your problem". Popularity/volume washes out of the equation, if mobile parts are really popular driving down the cost of raw wafers then Threadripper uses the same raw wafers so benefits just as much. In terms of yield, 9% extra area sounds pretty significant to me.

    And as I said, I don't believe for a second that Athlon X4 FM2 chips have faulty video circuitry. They are fused off and the GPU power budget aimed back at the CPU to create a new product.
    That assumes you can run the video decode part independently of the GPU,and whether IGPs have enough redundancy like larger dGPUs to tolerate loads of faults(which I doubt) and people need to accept that DIY PC building,gaming rigs,etc are not anywhere as large volume as bog standard APUs in laptops and prebuilt desktops. AMD server share is also tiny,and even doubling or tripling share,would still mean all those bog standard CPUs in laptops and desktop will sell more.

    Its a lower margin market,but its simply more important for AMD to offer a compelling product than the cheapest to make one,otherwise they could have easily just used a smaller IGP and made it smaller than Ryzen to build.

    There is nothing stopping AMD from sticking with 512 shaders which would still destroy Intel in the IGP wars,especially with the sigificantly better memory controller when compared to Bristol Ridge.



    Global Foundries has a history of not being able to do GPUs properly - again the X4 845 was a part which was not required at all. AMD had plenty of Steamroller based Athlon II X4 SKUs in the channel. Also this contradicts what people keep saying about the GPUs not being faulty. AMD offer loads of Athlon II X4 variants which indicates to me they were trying their best to offload APUs with faulty GPUs.

    The only reason for AMD to validate an Excacator based part in a sea of Steamroller based ones on FM2+ was because they obviously had enough Carrizo APUs with GPU faults to warrant the effort. If that wasn't the case with the X4 845 would never be released.

    Going back to Llano - one of the reasons it was delayed was due to poor yields on the GPU section:

    http://www.fudzilla.com/news/process...rom-bad-yields

    If anything the fact that Athlon II X4 CPUs exist indicates GPU issues. Why?? Look at the many of the APUs - they are unlocked.

    It makes little or no sense for AMD to not sell parts with faulty GPUs,as throwing them away would gain zero cash. OTH,there is little or no sense for them to offer a part with is 8C artificially disabled to 4C if they can - it would make more sense for them to use them for 6C and 8C variants which have a higher ASP,and then price it more competitively to take away sales from the Intel Core i5 CPUs.

    Edit!!

    Moreover,remember Ryzen+ probably will have some improvements to clockspeeds too,so AMD would rather offer a slightly more gimped Athlon II X4 with 4MB of L3 cache and lower clockspeeds(even when overclocked) than a salvaged Ryzen+ chip,ie,it will be better to upsell the 6C Ryzen+ CPUs.
    Last edited by CAT-THE-FIFTH; 10-01-2018 at 03:33 PM.


    Those despicable Elk,stealing the pond weed!

  8. #840
    Moosekateer CAT-THE-FIFTH's Avatar
    Join Date
    Aug 2006
    Location
    Not here
    Posts
    29,497
    Thanks
    3,318
    Thanked
    4,627 times in 3,571 posts
    • CAT-THE-FIFTH's system
      • Motherboard:
      • Less E-PEEN
      • CPU:
      • Massive E-PEEN
      • Memory:
      • RGB E-PEEN
      • Storage:
      • Not in any order
      • Graphics card(s):
      • EVEN BIGGER E-PEEN
      • PSU:
      • OVERSIZED
      • Case:
      • UNDERSIZED
      • Operating System:
      • DOS 6.22
      • Monitor(s):
      • NOT USUALLY ON....WHEN I POST
      • Internet:
      • FUNCTIONAL

    Re: AMD - Zen chitchat

    Raven Ridge based mini-PC spotted:

    https://www.computerbase.de/2018-01/...h-gemini-lake/
    https://imgur.com/a/1kxxO

    Edit!!

    Some more Ryzen and Ryzen Mobile systems are previewed:

    https://www.techpowerup.com/240471/a...ces-2018-booth

    A few more laptops,some AIO PCs and an Alienware gaming system!!
    Last edited by CAT-THE-FIFTH; 10-01-2018 at 05:10 PM.


    Those despicable Elk,stealing the pond weed!

  9. #841
    Senior Member Xlucine's Avatar
    Join Date
    May 2014
    Posts
    1,809
    Thanks
    244
    Thanked
    154 times in 121 posts
    • Xlucine's system
      • Motherboard:
      • Asus TUF B450M-plus
      • CPU:
      • 3700X
      • Memory:
      • 16GB @ 3.2 Gt/s
      • Storage:
      • Crucial MX500 1TB, Crucial MX100 512GB, 2TB hard disk
      • Graphics card(s):
      • EVGA 980ti
      • PSU:
      • Seasonic S12G-550
      • Case:
      • Silverstone TJ08-E
      • Operating System:
      • W10 pro
      • Monitor(s):
      • Viewsonic vx3211-2k-mhd, Dell P2414H
      • Internet:
      • Plusnet 70 Mb/s

    Re: AMD - Zen chitchat

    That mini PC looks nice, hopefully it undercuts the intel NUCs

  10. #842
    Not a good person scaryjim's Avatar
    Join Date
    Jan 2009
    Location
    Manchester
    Posts
    15,149
    Thanks
    1,224
    Thanked
    2,273 times in 1,865 posts
    • scaryjim's system
      • Motherboard:
      • Dell Inspiron
      • CPU:
      • Core i5 8250U
      • Memory:
      • 1x 8GB DDR4 2400
      • Storage:
      • 128GB M.2 SSD + 1TB HDD
      • Graphics card(s):
      • Radeon R5 230
      • PSU:
      • Battery/Dell brick
      • Case:
      • Dell Inspiron 5570
      • Operating System:
      • Windows 10
      • Monitor(s):
      • 15" 1080p laptop panel

    Re: AMD - Zen chitchat

    Quote Originally Posted by CAT-THE-FIFTH View Post
    ... especially with the sigificantly better memory controller when compared to Bristol Ridge. ...
    I've said it before CAT, that chart is either a misread or from a single-channel system. Here's Kaveri with dual channel DDR3-2133:


    source: http://hexus.net/tech/reviews/cpu/65...kaveri/?page=4

    There is no way that, moving from Kaveri to Bristol Ridge, AMD halved their memory controller bandwidth. That might be a write result, but AIUI write would be far less important then read in IGP performance. I don't buy that Bristol Ridge has a completely sub-par IMC.

    Anyway, your argument is kind of self-defeating. APUs are the high volume product, which proves that there's bound to be an APU derived CPU?

    As I said, the economics simply don't add up. AMD don't need a Raven Ridge derived CPU in this generation - they've got high performance CPUs already. With 11 total CUs, and only 3 needed for a functional Ryzen 3 2200U, there's scope to use more salvage dies for APU products, as they can tolerate 73% failure of the IGP. Deriving a CPU from Raven Ridge and pricing it at half the cost of a Summit Ridge CPU would ONLY hurt their own sales, and Summit Ridge is 10% more profitable at the same price as Raven Ridge.

    There would have to have appalling yield on the Raven Ridge IGP to even consider making CPUs out of them, and even then they'd have to be priced very close to the Summit Ridge CPUs to avoid tanking Summit Ridge sales. It makes no business sense.

    So basically, I'll be shocked (and a little horrified) if we get $50 quad core Ryzen CPUs. If AMD do make Raven Ridge derived CPUs, they'll cost about the same as the Raven Ridge APUs and be market-differentiated on something else (as I've said before, I'd put my money on energy efficiency).

  11. #843
    Moosekateer CAT-THE-FIFTH's Avatar
    Join Date
    Aug 2006
    Location
    Not here
    Posts
    29,497
    Thanks
    3,318
    Thanked
    4,627 times in 3,571 posts
    • CAT-THE-FIFTH's system
      • Motherboard:
      • Less E-PEEN
      • CPU:
      • Massive E-PEEN
      • Memory:
      • RGB E-PEEN
      • Storage:
      • Not in any order
      • Graphics card(s):
      • EVEN BIGGER E-PEEN
      • PSU:
      • OVERSIZED
      • Case:
      • UNDERSIZED
      • Operating System:
      • DOS 6.22
      • Monitor(s):
      • NOT USUALLY ON....WHEN I POST
      • Internet:
      • FUNCTIONAL

    Re: AMD - Zen chitchat

    Quote Originally Posted by scaryjim View Post
    I've said it before CAT, that chart is either a misread or from a single-channel system. Here's Kaveri with dual channel DDR3-2133:


    source: http://hexus.net/tech/reviews/cpu/65...kaveri/?page=4

    There is no way that, moving from Kaveri to Bristol Ridge, AMD halved their memory controller bandwidth. That might be a write result, but AIUI write would be far less important then read in IGP performance. I don't buy that Bristol Ridge has a completely sub-par IMC.
    It does have a subpar IMC - IIRC people like The Stilt were talking about it on AT forums. IIRC,they just licensed one from a third party company. This is why it can barely even run 2666MHZ memory when overclocked IIRC.

    #



    I pulled that off AT forums - both are running 2133MHZ memmory in dual channel. One is DDR3 and the other DDR4. The IMC was a POS in BR.


    Quote Originally Posted by scaryjim View Post
    Anyway, your argument is kind of self-defeating. APUs are the high volume product, which proves that there's bound to be an APU derived CPU?

    As I said, the economics simply don't add up. AMD don't need a Raven Ridge derived CPU in this generation - they've got high performance CPUs already. With 11 total CUs, and only 3 needed for a functional Ryzen 3 2200U, there's scope to use more salvage dies for APU products, as they can tolerate 73% failure of the IGP. Deriving a CPU from Raven Ridge and pricing it at half the cost of a Summit Ridge CPU would ONLY hurt their own sales, and Summit Ridge is 10% more profitable at the same price as Raven Ridge.

    There would have to have appalling yield on the Raven Ridge IGP to even consider making CPUs out of them, and even then they'd have to be priced very close to the Summit Ridge CPUs to avoid tanking Summit Ridge sales. It makes no business sense.

    So basically, I'll be shocked (and a little horrified) if we get $50 quad core Ryzen CPUs. If AMD do make Raven Ridge derived CPUs, they'll cost about the same as the Raven Ridge APUs and be market-differentiated on something else (as I've said before, I'd put my money on energy efficiency).
    The Llano based Athlon II X4 only happened due to poor IGP yields(check link in a previous post I made). Kaveri had production issues and was delayed massively with the GPU having the largest clockspeed missed,ie,they had issues there.

    Kaveri is actually slightly smaller than Carrizo/Bristol Ridge and yet why did you think AMD felt the need to release the Athlon II X4 845,when AMD has the Athlon II X4 830,840,860K,870K and 880K in the channel??

    You do realise some of the earlier single module chips with 128 shaders were based on the same Kaveri die?? Yet AMD also had tons of Athlon II X4 SKUs,which says to me they had enough APUs with totally dud IGPs.

    Oh wait,the GPU was faulty ON A MATURE 28NM PROCESS. The fact that AMD had so many Athlon II X4 models from the Steamroller line together with FX CPUs prove they had many Steamroller based APUs with faulty GPUs.

    Then included Polaris which has multiple revisions before launch and obviously missed its TDP targets. Its simply silly to ignore history to think GF has not had issues producting sufficient numbers of AMD APUs without IGP defects - they could not even produce a fully enabled Polaris 11 in enough quantity.

    Every APU launch since Llano has large numbers of Athlon II X4 CPUs available even when they had a competing platform,ie,AM3+ for example.

    AMD sells unlocked APUs too. There is no need for AMD to sell an Athlon II X4 in the first place,but only if the IGP is faulty.

    People are falling into the trap of thinking DIY sales of CPUs are the most of all CPU sales to consumers. They are not - hence why Intel has IGPs on their entire line of consumer CPUs outside their repurposed server ones.

    AMD historically has repurposed their server chips lower down the stack than Intel has - even a Core i7 8700K is a smaller chip than an Ryzen 7 1800X. It makes no sense for them to gimp a fully enabled chip in the first place to anything other than a Ryzen 5 6C/12T CPU.

    Whereas an RR CPU with a dud IGP would normally been thrown away - this is why the Athlon II X4 exists.

    The APUs are more important for AMD consumer sales than their CPUs longterm - the APUs are what will make or break them in the laptop market and for so many prebuilt PCs under £500,it will be most CPUs with IGPs too from the big brand companies.

    So by that extension as eventually the volume of APUs produced increases,together with the GF track record with producing APUs,ie,lots of wasted dies even on mature 32NM and 28NM processes,means that an Athlon II X4 part will be seen at some point,and it will be far more profitable for AMD to sell those parts to the DIY market than disabling half a Ryzen+ CPU.

    Plus you keep saying it "will hurt their own sales" - no it won't.

    They already have hurt sales of the smaller Ryzen 3 1200 and Ryzen 5 1400 by selling a faster Ryzen 3 2200G and 2400G at the same price point with an IGP,and the slightly larger die.

    It seems AMD thinks its better to do this - they obviously think its better destroying their sub £150 Ryzen sales in favour of the APUs.

    You have also forgotten Ryzen+ too. I suspect when that hits,AMD will quietly clear out Ryzen at the lower end of the market in favour of RR based APUs and at some point the Athlon II X4.

    Most likely it will have higher clockspeeds and other improvements. It makes entire sense to do that - if you have a 14NM £80 4C/4T RR based CPU which you want for a gaming build,if the 12NM £160 Ryzen 5 2600 is clocked higher,has three times the threads and overclocks better,there is potential to upsell it more,than having a whole range of cheaper Ryzen+ derived models in amongst the APUs which overclock to a similar level.

    Why should they artificially disable an 8C/16T CPU to 4C/4T when they can use it to lower the bulk pricing of said CPUs to OEMs and push Intel out of higher tier sales??

    It makes more sense for them to sell faulty APUs to the DIY market - where are the going to sell such CPUs?? OEMs won't want them,so its better they sell them in the DIY market.

    I mean you go on about the slightly larger APU die - if it was that big a deal AMD would never have released Raven Ridge for DIY builders since it is replacing slightly smaller Ryzen CPUs at similar price-points.

    However,they still think it makes them more money doing it then sticking with Ryzen. Its all to do with numbers - they probably will eventually make more APUs.

    This is why AMD stopped bothering with improving the FX line,and why they bothered spending money on the APUs.

    Those consumer line CPUs might not have the margins of a TR CPU,but they certainly still ship in enough volume to be very important for the bottomline of AMD. If anything those APUs and how well they can penetrate the laptop market and prebuilt PCs are really what will make or break AMD.

    Plus another thing none of you can answer - considering APUs will be mostly found in cheaper systems on desktop and even laptops,why do you think AMD went with such a "large" die. 512 shaders with faster RAM,a better memory controller and higher clockspeeds would still beat Intel easily for the next year or so. That would easily drop die size below Ryzen.

    It says to me they would rather have slightly higher production costs for the APUs since they think it worth going with a bigger IGP.

    Anyway,I don't think I am going to agree with you or DWL about this so we need to agree to disagree.
    Last edited by CAT-THE-FIFTH; 10-01-2018 at 09:52 PM.


    Those despicable Elk,stealing the pond weed!

  12. #844
    Not a good person scaryjim's Avatar
    Join Date
    Jan 2009
    Location
    Manchester
    Posts
    15,149
    Thanks
    1,224
    Thanked
    2,273 times in 1,865 posts
    • scaryjim's system
      • Motherboard:
      • Dell Inspiron
      • CPU:
      • Core i5 8250U
      • Memory:
      • 1x 8GB DDR4 2400
      • Storage:
      • 128GB M.2 SSD + 1TB HDD
      • Graphics card(s):
      • Radeon R5 230
      • PSU:
      • Battery/Dell brick
      • Case:
      • Dell Inspiron 5570
      • Operating System:
      • Windows 10
      • Monitor(s):
      • 15" 1080p laptop panel

    Re: AMD - Zen chitchat

    Quote Originally Posted by CAT-THE-FIFTH View Post
    ... Anyway,I don't think I am going to agree with you or DWL about this so we need to agree to disagree.
    Well, that's up to you. For me there's three things:

    1) The write bandwidth is the issue with the construction core IMC, by the look of it - the read and copy are pretty much where you'd expect. I don't actually know how much impact that would have, tbh!

    2) you've linked one story that says the Llano IGP was yielding poorly, but Trinity was doing much better. After Richland the APUs moved on to 28nm, and if you're honestly expecting me to believe that AMD, after a couple of years of fabbing GPUs on 28nm, were still having problems fabbing an IGP on 28nm ... well, I personally think that's unlikely. It's certainly a long way from clear proof that the Athlon X4 chips existed solely because AMD had a bunch of dies with completely inoperable GPUs...

    3) Let's turn this around a bit - there are currently 6 known variants of the Raven Ridge die - 4C/8T with 11, 10 or 8 CUs, 4C/4T with 8 or 6 CUs, and 2C/4T with 3CUs. Which of those do you imagine would become this theoretical Athlon, and what might it cost?

    It looks like the desktop Raven Ridge is overclockable, and APUs are already being sold cheaper than similarly configured CPUs. I honestly don't see where a Raven Ridge based CPU would fit into the product stack. There's no sensible space. The 2200G and 2400G have replaced the 1200 and 1400. If a Raven Ridge CPU were launched, it would most like replace the existing CPUs. It looks far more likely that we'll see a rotating release of CPUs and APUs - there'll be 2300X and 2500X CPUs cut from 12nm Ryzen 2, which will replace the 1300X and 1500X. They won't leave space for RR CPUs, because who'd buy a 2300X if there's a 4-core RR CPU selling cheaper? The positioning of the 2200G and 2400G strongly suggest that AMD's product stack will involve an APU and a CPU for each product configuration, and that the CPU will be faster and more expensive than the APU. The only space to fill - if they choose to do so - is the 2C/4T entry level at around $50. It's yet to be seen if they'll bother with a CPU-only for that market.
    Last edited by scaryjim; 10-01-2018 at 11:39 PM.

  13. #845
    Not a good person scaryjim's Avatar
    Join Date
    Jan 2009
    Location
    Manchester
    Posts
    15,149
    Thanks
    1,224
    Thanked
    2,273 times in 1,865 posts
    • scaryjim's system
      • Motherboard:
      • Dell Inspiron
      • CPU:
      • Core i5 8250U
      • Memory:
      • 1x 8GB DDR4 2400
      • Storage:
      • 128GB M.2 SSD + 1TB HDD
      • Graphics card(s):
      • Radeon R5 230
      • PSU:
      • Battery/Dell brick
      • Case:
      • Dell Inspiron 5570
      • Operating System:
      • Windows 10
      • Monitor(s):
      • 15" 1080p laptop panel

    Re: AMD - Zen chitchat

    I've been doing some thinking about market position and profitability for AMD.

    Bristol Ridge has a die size (albeit on a theoretically cheaper 28nm node) of ~ 250 mm2, but because of the performance position they could only charge ~ £100 for the top-end fully enabled product based on that die. That can't have been very profitable

    With Raven Ridge, a 210mm2 die, they are selling the fully enabled die for ~£150. They're essentially giving away the IGP at that price (the CPU is almost identical to the similarly priced 1st Gen CPUs), so presumably they've set that price at a profitable level - they could easily justify charging more for the IGP since they have CPU-only parts in the market as well, but they've taken a business decision to charge basically the same whether you want a CPU only or a CPU+IGP.

    But at that price, they're selling the fourth or fifth binning of Summit Ridge, which is a smaller die on the same process. In pure cost of silicon terms, just how profitable is Summit Ridge?! Assuming similar yields and packaging costs, a £150 Summit Ridge CPU should be slightly more profitable than a £150 Raven Ridge APU, and that's the top binning of Raven Ridge against a much lower binning of Summit Ridge (remember that the 1800X isn't the top binning - the best dies go in to Threadripper).

    It also explains why they can afford to cut Raven Ridge all the way down to a 2C/3CU high volume mobile part while they couldn't with Bristol Ridge - they can probably charge as much for that as for a 2nd or 3rd bin quad core Bristol Ridge, so I wouldn't be surprised if they were still turning a profit on those chips, even cut down that heavily. And it suggests that - if it ever materialises in consumer products - a separate 2C/3CU die would be stupidly cheap.

  14. #846
    root Member DanceswithUnix's Avatar
    Join Date
    Jan 2006
    Location
    In the middle of a core dump
    Posts
    10,808
    Thanks
    577
    Thanked
    1,160 times in 986 posts
    • DanceswithUnix's system
      • Motherboard:
      • Asus X470-PRO
      • CPU:
      • 3700X
      • Memory:
      • 16GB 3200MHz
      • Storage:
      • 1TB Linux, 1TB Games (Win 10)
      • Graphics card(s):
      • Asus Strix RX Vega 56
      • PSU:
      • 650W Corsair TX
      • Case:
      • Antec 300
      • Operating System:
      • Fedora 30 + Win 10 Pro 64 (yuk)
      • Monitor(s):
      • Benq XL2730Z 1440p + Samsung 2343BW 2048x1152
      • Internet:
      • Zen 80Mb/20Mb VDSL

    Re: AMD - Zen chitchat

    Quote Originally Posted by scaryjim View Post
    I've been doing some thinking about market position and profitability for AMD.
    Is this still debate going?

    Some snippets to mix in to your observations which I think are all correct but can be added to.

    With BR, AMD's wafer agreement and low sales volume means they were going to pay for those wafers whether they make them or not, so making poor/no profit parts is better than a flat loss (and maintains at least some market share)

    As well as the silicon costs, you have to factor in the fixed design costs for the core which would have been pretty significant. Summit Ridge coming out first would have made AMD keen to get a return on that design cost which pushes the part cost up. When Raven Ridge came out, the CPU core design hasn't changed so no extra engineering costs there, but it might have been decided that either/and:

    1/ Enough money was made on SR that RR doesn't need to claw back as much of that module design cost
    2/ RR will probably sell in higher volume, so you get more devices to amortise the fixed cost over compared to SR

    So yes "In pure cost of silicon" terms you are right that SR is way more profitable, but the accountants might say that is unfair.

    In terms of dramatic cut down parts, cast a mind back to the Athlon X2 5000+. Mine unlocked to a Phenom II X3, and I get the impression I was unlucky. Business reasons tend to trump technical ones.

  15. #847
    Not a good person scaryjim's Avatar
    Join Date
    Jan 2009
    Location
    Manchester
    Posts
    15,149
    Thanks
    1,224
    Thanked
    2,273 times in 1,865 posts
    • scaryjim's system
      • Motherboard:
      • Dell Inspiron
      • CPU:
      • Core i5 8250U
      • Memory:
      • 1x 8GB DDR4 2400
      • Storage:
      • 128GB M.2 SSD + 1TB HDD
      • Graphics card(s):
      • Radeon R5 230
      • PSU:
      • Battery/Dell brick
      • Case:
      • Dell Inspiron 5570
      • Operating System:
      • Windows 10
      • Monitor(s):
      • 15" 1080p laptop panel

    Re: AMD - Zen chitchat

    Quote Originally Posted by DanceswithUnix View Post
    Is this still debate going?
    No, it's more just something that was running round my head while I was trying to get to sleep last night. Now it's no longer in my head, maybe I'll get more sleep tonight!

    With BR, AMD's wafer agreement and low sales volume means they were going to pay for those wafers whether they make them or not, so making poor/no profit parts is better than a flat loss
    Yeah, I kind of assumed that they were at best selling at cost for the dual-core desktop parts, which were all less than £50 iirc. I also noticed while looking around that the Kaveri & Carrizo x4 CPUs sold for more than the dual core APUs. Since the lowest CU count for a quad core APU was 6, a part with 5 working CUs and 4 working cores was more profitable to sell as a CPU than an APU. As to the X4 845, since Carrizo APUs were mobile only they needed a product to bag off any parts that wouldn't run at low voltages. Releasing an entire line of APUs would mean expensive binning and marketing, so why not just slap everything with 4 working cores that mnissed the voltage target into the Athlon line...

    So yes "In pure cost of silicon" terms you are right that SR is way more profitable, but the accountants might say that is unfair.
    Yeah, granted. This was more of a musing because the side-by-side comparison of RR and SR suddenly made it hit home to me that the original selling price of the 1800X must have had a phenomenal profit margin. It certainly explains how Intel and Nvidia manage those massive net margins.

    ... In terms of dramatic cut down parts, cast a mind back to the Athlon X2 5000+. Mine unlocked to a Phenom II X3, and I get the impression I was unlucky. Business reasons tend to trump technical ones. ...
    Yeah, Uriel is still updating his Phenom FX5200 thread through seven years of running a dual core Athlon as a quad core Phenom. It's the most obvious example I can think of of perfectly good silicon being fused off to form a new product line, although there was a LOT of core unlock capacity in the whole Phenom II derived line (after all, my Sempy happily unlocked to a dual core Athlon).

  16. #848
    root Member DanceswithUnix's Avatar
    Join Date
    Jan 2006
    Location
    In the middle of a core dump
    Posts
    10,808
    Thanks
    577
    Thanked
    1,160 times in 986 posts
    • DanceswithUnix's system
      • Motherboard:
      • Asus X470-PRO
      • CPU:
      • 3700X
      • Memory:
      • 16GB 3200MHz
      • Storage:
      • 1TB Linux, 1TB Games (Win 10)
      • Graphics card(s):
      • Asus Strix RX Vega 56
      • PSU:
      • 650W Corsair TX
      • Case:
      • Antec 300
      • Operating System:
      • Fedora 30 + Win 10 Pro 64 (yuk)
      • Monitor(s):
      • Benq XL2730Z 1440p + Samsung 2343BW 2048x1152
      • Internet:
      • Zen 80Mb/20Mb VDSL

    Re: AMD - Zen chitchat

    Quote Originally Posted by scaryjim View Post
    Releasing an entire line of APUs would mean expensive binning and marketing, so why not just slap everything with 4 working cores that mnissed the voltage target into the Athlon line...
    There were rumours at the time that the video section just didn't work on AM2+ motherboards.

    AMD claimed that the 845 was all about power consumption, which would tie in with the Athlon X2 5000+ where fusing off most of a top end 45nm Phenom II produced a lower power consuming device than the actual dual core of the time (which I guess must have been 65nm but am too lazy to fact check on a Friday lunchtime ).

    Yeah, granted. This was more of a musing because the side-by-side comparison of RR and SR suddenly made it hit home to me that the original selling price of the 1800X must have had a phenomenal profit margin. It certainly explains how Intel and Nvidia manage those massive net margins.

    Yeah, Uriel is still updating his Phenom FX5200 thread through seven years of running a dual core Athlon as a quad core Phenom. It's the most obvious example I can think of of perfectly good silicon being fused off to form a new product line, although there was a LOT of core unlock capacity in the whole Phenom II derived line (after all, my Sempy happily unlocked to a dual core Athlon).
    People are kind of used to getting milked on high end parts. I think the only time I ever had a top of the line CPU was an Athlon 1300 which at the time was about £150, but then there was genuine competition with Intel at that point.

    Edit: Hope you sleep better for thinking this through

  17. Received thanks from:

    scaryjim (12-01-2018)

Thread Information

Users Browsing this Thread

There are currently 3 users browsing this thread. (0 members and 3 guests)

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •